CN102538717B - Automatic leaf area index observation system and method thereof - Google Patents

Abstract

A kind of automatic leaf area index observation system and method thereof, described system comprises: data acquisition unit; And leaf area index automatic Observation server, it comprises: bianry image generation module, parameter extraction module, leaf area index computing module, wherein bianry image generation module is categorized into bianry image for the crop canopies digital picture that data acquisition unit is obtained, described parameter extraction module is for extracting clearance rate, gap size, aggregate index parameter from bianry image, and described leaf area index computing module is for calculating leaf area index according to the information parameter extracting. System of the present invention has realized leaf area index indirect measurement method. Than prior art, the present invention can be used for field Remote Acquisitioning picture, meets the Images Classification of field complex condition, and nicety of grading is reliable, and classification results is more accurate. From information such as bianry image statistics clearance rate, gap size and aggregate index, can carry out more deep analysis, calculate leaf area index.

Description

Automatic leaf area index observation system and method thereof

Technical field

The present invention relates to agriculture observation field, more specifically, relate to automatic leaf area index observation system and method thereof.

Background technology

Leaf area index (LAI, LeafAreaIndex), as one of conventional agroecological environment parameter of fundamental sum, becomes one of important index in fields such as phytoecology, forestry and agricultural estimation output. At present temperature, humidity, the soil water parameter of grading has ready-made sensor, but also there is no leaf area index sensor, and this is because the particularity that leaf area index is measured causes. In general, under the condition of not damaging crop, be difficult to directly measure leaf area index, if with indirect measurement method, need to solve the problem typical of sampling, automatic distinguishing blade and background, and to leaf inclination angle and the disturbing factor such as mutually block and correct. Because these difficult points, the instrument of existing indirect measurement leaf area index all needs manual operation, can't accomplish full automation. Still there is no at present automatic leaf area index observation system.

Summary of the invention

For the problems referred to above, the present invention proposes a kind of automatic leaf area index observation system and method thereof, thereby under the prerequisite of limited cost, solve a difficult problem for all automatic measurement leaf area index.

Described automatic leaf area index observation system, comprising:

Leaf area index automatic Observation server, comprise: bianry image generation module, parameter extraction module, leaf area index computing module, wherein bianry image generation module is categorized into bianry image for the crop canopies digital picture that data acquisition unit is obtained, described parameter extraction module is for extracting clearance rate, gap size, aggregate index from bianry image, and described leaf area index computing module is for calculating leaf area index according to the information parameter extracting;

Data acquisition unit, comprise: for the solar energy intelligent electricity generation system of powering, gather the camera sensing device of crop digital picture, for transmitting the digital transmission module of data and for receiving the data of collection and being transferred to the data acquisition board of described leaf area index automatic Observation server by output module.

Described leaf area index automatic observation process, comprising:

The crop canopies digital picture that data acquisition unit is obtained is categorized into bianry image;

Extract clearance rate, gap size, aggregate index from bianry image;

Calculate leaf area index according to the information parameter extracting.

Automatic leaf area index observation system of the present invention is to design for extract the long-pending index of corps leaf surface from digital picture, and it has realized leaf area index indirect measurement method. Than existing similar system as CANEYE, CIMES-FISHEYE etc., these two systems are all to design for fisheye camera, system of the present invention is to process and design for general camera digital image, this discovery simultaneously provides 3 kinds of automatic classification methods, can meet the Images Classification of field complex condition. 3 kinds of sorting techniques are all independent propositions, reliable through experimental verification nicety of grading. System of the present invention mainly comprises bianry image generation module, parameter extraction module and leaf area index computing module. Images Classification is normally divided into crop canopies digital image Soil Background and green vegetation two classes. Improved K-average sorting technique is first divided into multiclass automatically by image, then carries out threshold process according to all kinds of wave band averages, merges into green vegetation and soil and background two classes. Automatic threshold classification comprehensive utilization RGB and HSL color space, provide 4 kinds of wave band threshold value combined methods. Divide block threshold value classification to carry out piecemeal processing for the more roomy image of maize leaf or the more roomy crop original image of other blades, after piecemeal, original image becomes uniform figure spot, then utilizes wave band threshold value to carry out binaryzation, and classification results is more accurate. Can add up the information such as clearance rate, gap size distribution curve and aggregate index from bianry image, utilize these information can carry out more deep analysis, thereby calculate leaf area index.

Brief description of the drawings

Fig. 1 is the structural representation of automatic leaf area index observation system of the present invention;

Fig. 2 is the scheme of installation of the data acquisition unit of automatic leaf area index observation system of the present invention;

Fig. 3 is the circuit diagram of the data acquisition board of automatic leaf area index observation system of the present invention;

Fig. 4 is the flow chart of leaf area index automatic observation process of the present invention;

Fig. 5 is the schematic diagram of RGB of the present invention color space and HSL color space;

Fig. 6 is the schematic diagram of bianry image generation module of the present invention;

Fig. 7 is the example of the original image of collection of the present invention;

Fig. 8 be shown in the example of the bianry image that generates after classification of the present invention of original image;

Fig. 9 is the schematic diagram that gap size of the present invention distributes;

Figure 10 is the example that Estimating Leaf Area In Maize index extracts result.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, the present invention is further detailed explanation.

Fig. 1 is the structural representation of automatic leaf area index observation system of the present invention, and as seen from the figure, area index automatic observing system of the present invention comprises data acquisition unit and leaf area index observation server. Data acquisition unit is arranged in farmland to be observed, and the data of its observation send long-range leaf area index calculation element to by network, calculates leaf area index by leaf area index observation server.

In practice, data acquisition unit is arranged on support bar, and post upper has towards the support lever arm of a side, and support lever arm is overhead highly preferred is 3 meters. Certainly, also can regulate according to the height of the crops that detect the setting height of support lever arm.

Described data acquisition unit comprises: solar energy intelligent electricity generation system, data acquisition board, camera sensing device, Temperature Humidity Sensor, GPS sensor, digital transmission module.

Wherein, solar energy intelligent electricity generation system is made up of solar panels, battery and charging control unit; Battery is used to described data acquisition unit power supply; It is charge in batteries that charging control unit is used for controlling solar panels. Described solar panels are accepted the radiation of sunshine daytime, and transform light energy is become to electric energy, and charge to battery 8 under the effect of charging control unit, meanwhile, in the time that solar energy is sufficient, power directly to described device by solar panels. For example: it is 19V that solar panels provide upper voltage limit value, in the time that solar panels voltage exceedes battery tension, by automatic accumulators power supply; When solar energy is sufficient, power to described data acquisition unit by solar panels; When solar energy is not enough, power to described data acquisition unit by battery; When battery is during lower than 10V, described data acquisition unit auto sleep, to protect the life-span of battery, until solar energy intelligent electricity generation system is when starting next time, solar panels charge a battery and exceed 10V.

Wherein, adopt two camera sensing devices, it is fixed on support lever arm external end head, and observation angle is respectively vertically downward (zenith angle 0 is spent) and 57.7 degree that tilt, and visual field 60 is spent.

Camera sensing device BR, gathers crop canopies digital image information, and transfers to data acquisition board. What camera sensing device adopted is 5,000,000 high-definition digital cameras, and resolution ratio is 320*240～2596*1944, and resolution ratio when camera sensing device gathers can arrange as required. The automatic environment-identification light intensity of camera sensing device, when insufficient light, will in the time taking, open flash lamp, the normal need of work 12V power supply of module, by serial ports and data acquisition board communications, only power-on switch in the time receiving shooting order, thus power supply saved. The time of taking can be according to relevant parameter setting, within one day, can take at most 12 times, and system is according to setting-up time pictures taken. For example: twice of 7:00 and 16:00 automatic shooting in afternoon every morning; In the time meeting with rainy weather, according to circumstances select pot life point data. In the middle of one day, gather four width images, select the image that the quality of data is higher be used for classification and can utilize two angular image results to be averaged from four width images, finally export a leaf area index value. The image that camera sensing device is taken is stored in the memory of signal acquiring board, if insufficient memory capacity or other faults, picture will can not be stored. If for example, all cannot normally work at the scheduled time (1 point of half, wherein half a minute is that camera powered on until the stable time needing) interior camera, will stop shooting task.

Wherein, Temperature Humidity Sensor is for collecting temperature, humidity information and be transferred to data acquisition board, and Temperature Humidity Sensor is welded in data acquisition board. For example can adopt sht10 digital sensor, its operating voltage 3.3V, only in the time of sampling just to its power supply.

Wherein, GPS sensor, for automatically revising the RTC real-time clock of data acquisition board, records coordinate, opens at the zero point of every day, and all the other times, GPS sensor was welded in data acquisition board in closed condition. Can set GPS sensor opened at the zero point of every day, find satellite, and carry out time adjustment to data acquisition board finding after satellite, and residing system longitude and latitude is recorded, be transferred to long-range leaf area index observation server with image data by digital transmission module as described below.

Wherein, digital transmission module is connected in data acquisition board, near data acquisition board, for the data message of data acquisition board (comprising the data of memory, the humiture of collection and system voltage, latitude and longitude information) is transferred to long-range leaf area index calculation element. Digital transmission module can utilize existing wireless network or cable network to carry out transfer of data. Digital transmission module can be selected 2.4G module, iridium satellite digital transmission module, radio module and GPRS module. This description is described the present invention as an example of GPRS module example, and those skilled in the art can expect other replacement embodiment under instruction of the present invention. Mobile phones universal packet radio service technology (GPRS) can be realized data wireless transmission, adopt 3.8V power supply, in the time that completing, shooting task opens at every turn, also can carry out opening and logging at each hour integral point, this is in the situation of GPRS network poor signal, can ensure to connect and connect to the Net smoothly, ensure the remote transmission of data. Digital transmission module transmits the data in the memory in data acquisition board, the complete automatic closing module power supply of transfer of data after connecting remote server.

Wherein, the information that data acquisition board gathers for gathering above-mentioned each parts, and observe server carry out transfer of data by digital transmission module and long-range leaf area index. Data acquisition board comprises microcontroller and deposits all devices, more specifically, microcontroller is used for gathering the information of Temperature Humidity Sensor collection, the crop canopies digital image information of camera sensing device output, the information of voltage that solar energy intelligent electricity generation system sends, the time adjustment information that GPS sensor gathers, and for controlling the unlatching of solar energy intelligent electricity generation system, camera sensing device, Temperature Humidity Sensor, GPS sensor and digital transmission module. Deposit the information that all devices gather for storing microcontroller. Wherein, holder can be selected the nandflash of 1G, and the information of collection is kept in nandflash, also can read the information of having preserved from nandflash, if after information reads, will automatically delete, for new below information storage. Data acquisition board is contained in a seal box, prevents from being subject to weather effect. Seal box is fixedly connected on support bar middle part.

Fig. 2 has shown the figure that installs and uses of described data acquisition unit. Fig. 3 has shown the circuit diagram of data acquisition board.

Automatic leaf area index observation system of the present invention also comprises leaf area index observation server, it is communicated by letter with described data acquisition unit by internet, and described leaf area index observation server comprises bianry image generation module, parameter extraction module, leaf area index computing module. Bianry image generation module is categorized into bianry image for the crop canopies digital picture that data acquisition unit is obtained. Described parameter extraction module is for extracting the parameters such as clearance rate, gap size, aggregate index from bianry image. Leaf area index computing module is for calculating leaf area index according to the information parameter extracting.

Describe the course of work of described leaf area index observation server in detail below with reference to Fig. 4.

Method of the present invention has been utilized RGB and the HSL color space in Digital Image Processing, and Fig. 5 has shown respectively the schematic diagram of these two color spaces. In RGB color space, true color image can be decomposed into red R, green G and tri-components of blue B, the value of each component is 0-255, (0,0,0) represent that black, (255,255,255) represent white, (255,0,0) represent that red, (0,255,0) represents green, (0,0,255) represent blueness. Classification foundation of the present invention is exactly crop leaf for green, just can distinguish green vegetation and background by extracting green pixel.

In HSL color space, true color image can be decomposed into colourity H, saturation degree S and tri-components of brightness L, H value is 0-360 degree, S value 0-1, and L value is 0-1. Classification foundation of the present invention is that crop leaf is green, and corresponding H span is 60-180, and its value also may be subject to the impact of image-forming condition, so user can carry out threshold value setting or select the threshold value of recommending according to picture quality.

1, classification generates bianry image

In this process, be categorized into bianry image by bianry image generation module for the crop canopies digital picture that data acquisition unit is obtained.

Images Classification is normally divided into crop canopies digital image Soil Background and green vegetation two classes. Experiment shows, due to crop species difference, or plant growth stage difference, image classification method used is not identical yet, there is no a kind of pervasive sorting technique. For example, be winter wheat and corn North China chief crop, after winter wheat harvesting, stalk is stayed field more, and soil and stalk mix the complexity that has increased classification. Should select different sorting techniques for winter wheat and corn, also need to change sorting technique for corn at different growth phases. To this, the invention provides three kinds of sorting techniques: improved K-average classification, automatic threshold classification, point block threshold value classification. Realized by the improved K-average sort module in the bianry image generation module shown in Fig. 6, automatic threshold sort module and a point block threshold value sort module respectively.

The first embodiment: improved K-average sorting technique

The method is realized by the improved K-average sort module in bianry image generation module.

Improved K-average sorting technique has been used clustering method, need in data, select required classification number, the random cluster centre position of searching, then iteration reconfigure them, until reach optimized classification. Improved K-average sorting technique is first divided into multiclass automatically according to selected classification number, then carries out threshold process according to all kinds of wave band averages, merges into green vegetation and soil and background two classes. The handling process of improving one's methods is:

1) equally distributed K initial category average on computed image data space;

2) calculate pixel to the distance of each initial category, they are gathered in nearest class by beeline method;

3) each iteration recalculates the average of classification, and pixel is classified again by new classification average;

4) iterative step 2～3 is until meet set condition, if define error threshold, in the time that each class pixel number that classification changes in an iteration is less than threshold value, iteration stops. Reach error threshold or reach greatest iteration and now classify and all will finish;

5) than primal algorithm, all kinds of averages in RGB color space when improved K average statistic of classification classification stops, be merged into a class if meet the classification of G > R and G > B, represent green blade, remaining classification is merged into Soil Background class, finally realizes the binaryzation of image by this step.

The second embodiment: automatic threshold sorting technique

The method is realized by the automatic threshold sort module in bianry image generation module.

For winter wheat and the less image of maize leaf, provide four kinds of wave band threshold value sorting techniques. R, G, B represent respectively three wave bands of red, green, blue of RGB color space, their span is 0-255, H represents the colourity in HSL color space, its span is 0 °-360 °, the principle of classification is to distinguish green blade and Soil Background by R, G, B and tetra-ranges of variables of H are set, therefore define four threshold value t1, t2, t3, t4 and be used for arranging R, G, B and tetra-ranges of variables of H, in each method, the variable of four threshold value restrictions is fixing, is only the use to four variable assignments. Four kinds of methods provide the threshold value of recommending below, need to select suitable sorting technique according to image situation in application.

Method 1:(t1 < H < t2) or (R > t3 and G > t3 and B > t3) t1=60, t2=180, t3=200

In method 1, condition 1,60 < H < 180, the just in time scope of corresponding green blade H component in HSL color space; Condition 2, brighter blade-section in (R > 200 and G > 200 and B > 200) correspondence image, two conditions are got union and can obtain the classification results of degree of precision.

Method 2:(G > R+t1 and G > B+t2) or (R > t3 and G > t3 and B > t3) t1=0, t2=0, t3=200

In method 2, condition 1, (G > R and G > B) be the feature of corresponding green blade R, G, tri-components of B in rgb color space just in time; Condition 2, brighter blade-section in (R > 200 and G > t200 and B > 200) correspondence image, two conditions are got union and can obtain the classification results of degree of precision.

Method 3:(t1 < H < t2) and (G > R+t3 and G > B+t4) t1=60, t2=180, t3=0, t4=0

In method 3, the just in time scope of corresponding green blade H component in HSL color space of condition 1,60 < H < 180; Condition 2, brighter blade-section in (G > R+t3 and G > B+t4) correspondence image, t4 is set as 0 value conventionally, and user can, according to picture quality setting threshold voluntarily, keep flexibility.

Method 4:(G > R+t1 and G > B+t2) or ((R-G) < t3) or (R > t4 and G > t4 and B > t4) t1=0, t2=0, t3=10, t4=200

Compared with method 2, method 4 has increased criterion ((R-G) < t3), this mainly designs for the blade solar flare in image, the R in RGB color space and the G component of finding blade solar flare through experiment meet such condition, and user can be according to picture quality setting threshold.

The 3rd embodiment: point block threshold value sorting technique

The method is realized by point block threshold value sort module in bianry image generation module.

Preferably, the more roomy image of maize leaf or the more roomy crop original image of other blades are carried out to piecemeal processing, it is more accurate that original image becomes uniform figure spot classification results, then utilizes wave band threshold value to carry out binaryzation. Piecemeal diameter determines the size of block diagram spot, and user need set according to image.

The handling process of this method is:

1) image is cut apart, and image is cut apart the diameter and the iterations that need user to input block, has cut apart rear image and has generated uniform figure spot, often has unassignable pixel on figure spot border. Cut apart rear image more even, classify more accurate.

2) assignment pixel does not merge, and calculates in assignment pixel neighborhood not average as the value of this pixel.

3) threshold value classification adopts the second sorting technique, and the threshold condition of recommendation is

(G > R+t1 and G > B+t2) or ((R-G) < t3) or (R > t4 and G > t4 and B > t4) t1=0, t2=0, t3=10, t4=200

By above three kinds of sorting techniques, crop canopies digital picture can be divided into Soil Background and green vegetation two classes, convert bianry image to.

Fig. 7 is the original maize canopy image receiving, and Fig. 8 is the bianry image that classification obtains afterwards, and wherein white represents green corn, and black represents interstices of soil. Contrast Fig. 7 and 8 can find out that original maize canopy image is divided into corn and Soil Background two classes accurately, it is worthy of note and adopt the planting patterns of straw-returning in the North China Plain, the background that stalk and soil mix has increased the complexity of classification, and experiment shows that the sorting technique precision that the present invention proposes can meet application requirements.

Preferably, before classification generates bianry image, also comprise pre-treatment step, the raw image data receiving is carried out to image processing, comprise image cropping, the geometry deformation of rectified print and illumination are non-homogeneous etc.

2, extract clearance rate, gap size, aggregate index.

Extract clearance rate, gap size, aggregate index by parameter extraction module for the bianry image generating.

The method of extraction clearance rate of the present invention, gap size, aggregate index and the below computational methods of leaf area index have close ties.

Leaf area index computational methods of the present invention comprise two kinds of methods: improved Lang&Xiang method and improved Chen&Chilar method.

Lang&Xiang leaf area index (LAILX) method referring to " A.R.G.Lang, yu-kin. calculate the leaf area index of discontinuous canopy from direct solar radiation transmitance, the research of [M] agroecological environment, Meteorology Publishing House ". It has discussed the reckoning problem that has leaf area index in the situation of large space in canopy theoretically with in experience. The key solving is that clearance rate is asked in segmentation in whole measuring route. This theoretical assumption is obeyed Poisson distribution at segmented paths blade, if Statistical Area becomes face, segmented paths is converted into cell.

The present invention improves Lang&Xiang method, automatically determines the size of image cutting unit, needs input observation zenith angle.

Lang&Xiang theoretical assumption is obeyed Poisson distribution at segmented paths blade, and Chen etc. think that this hypothesis in practice can not be completely satisfied, thinks that statistics void size for the raising of leaf area index estimation precision highly significant simultaneously. Therefore proposed Chen&Chilar leaf area index (LAICC) method, it has improved leaf area index optical instrument certainty of measurement based on vegetation canopy gap size analysis theories. Overlap theoretical Chen based on this and invented TRAC instrument, it can, for measuring the leaf area index (seeing " JingM.Chen; JosefChilar; Plantcanopygap-sizeanalysistheoryforimprovingopticalmeas urementsofleaf-areaindex, [J], AppliedOptics; 1995; 34 (27), 6211-6222 ") of non-homogeneous canopy, be used for the measurement of Forest Leaf Area Index in practical application in theory. Mostly the vegetation canopy of occurring in nature is gathering, under identical leaf area index condition, more may there is large scale space than the canopy of random distribution in the canopy of assembling, or can think and have a maximum interspace size that may occur for the canopy of random distribution, the space that exceedes so this size is just thought to be caused by building-up effect. These large-sized spaces have increased canopy clearance rate, and have affected the indirect measurement of leaf area index. Present problem is converted into the maximum interspace size of determining that random canopy occurs. Method of the present invention is optimized the method for removing large scale gap, has proposed improved Chen&Chilar leaf area index method.

Clearance rate (gapfraction) is the probability that direct sunlight sees through vegetation canopy arrival ground. It in digital image, is the ratio of black pixel in certain statistical regions (soil and background). Calculate and need clearance rate information for described below improvement Lang&Xiang leaf area index. Image is divided into little cell, adds up the clearance rate of each cell. From the data characteristics of bianry image, utilize square window statistics entire image more reasonable at the clearance rate of each cell.

The value of bianry image Green blade is 255, and the value of Soil Background is 0, from the formula of bianry image statistics clearance rate is:

$P\left(\mathrm{\&theta;}\right)=\frac{n{\mathrm{um}}_0}{i{\mathrm{mg}}_w\&times;{\mathrm{img}}_h}$

P (θ) represents clearance rate, img_wAnd img_hThe width and the height that represent respectively statistical picture, statistical picture can be that entire image can be also a region of cutting apart from image, their product representative image pixel sum, num₀The number of 0 value pixel in representative image.

Void size (gapsize) is that direct sunlight sees through the physical length of vegetation canopy arrival ground projected spot on direction of measurement. It in digital image, is the pixel count of continuous black pixel (soil and background) in a line, add up all image lines, then the number of times space of same size being occurred sums up, and just can obtain the frequency of the space appearance of this size at the pixel number divided by whole image. Need void size information for described improvement Chen&Chilar leaf area index below, improvement Chen&Chilar leaf area index is shown in operation in detail.

Aggregate index (clumpingindex) quantitative description the growth conditions of vegetation blade, be called as aggregate index because nature vegetation Leaf positional distribution is this parameter of coherent condition more, be numerically equal to effective leaf area index divided by true leaf area index. Provide respectively the computational methods of aggregate index separately for improvement Lang&Xiang leaf area index and improvement Chen&Chilar leaf area index below, operation is shown in improvement Lang&Xiang leaf area index (CILX) and improves Chen&Chilar leaf area index (CICC) in detail.

The invention provides two kinds of methods and extract clearance rate, gap size, aggregate index.

The first embodiment: based on improved Lang&Xiang method

After Lang&Xiang method supposition is less cell scene partitioning, the Leaf positional distribution in cell is more approaching to be uniformly distributed at random, and on whole area, the distribution of leaf has not just been subject to the constraint of this condition of random distribution. Therefore the LAI that utilizes simple logarithmic formula to calculate in each junior unit lattice approaches true LAI, to all cell calculating mean values as actual value, that calculate by the mean gap rate of whole scene is effective LAI, the ratio calculation aggregate index of the mean value of the true LAI that so just can calculate by effective LAI of whole scene and all junior unit lattice. Lang&Xiang leaf area index (LAILX) is called again limit for length's method of average, its hypothesis blade is in local random distribution, if statistics community area (path) equates, actual leaf area index equals the arithmetic mean of instantaneous value of all statistics community (path) leaf area index, and effective leaf area index can calculate with poisson's theorem.

The problem of Lang&Xiang method in application is the area Δ A that needs to determine statistics community, so that leaf distribution more approaching random (Poisson) in community is distributed. Because blade is the base unit of light interception body in canopy, so the gap distribution of canopy has positive bi-distribution feature, and the cell size that this positive bi-distribution feature is chosen at us is especially obvious while approaching blade dimensions. And along with the heteropical accumulation in large scale more, gap distribution trends towards negative binomial distribution gradually, the aggregate index that therefore we obtain is conventionally to be all less than 1 value. At this, we also propose a kind of method of definite division unit lattice size by the principle of error analysis, think if cell is too little, the larger error that can bring LAI to estimate, therefore for determining unit lattice splitting scheme must improve and potential error is tried to achieve balance between increasing in potential precision. The error of calculation of community leaf area index derives from leaf in community and departs from the error that Poisson distribution causes, and therefore the present invention utilizes the method for error analysis to determine community area Δ A.

After having told about our ratio juris, then state that in detail routine processes flow process is as follows:

Under each statistics community area, add up the clearance rate P of each community by cell_cell(θ) calculate aggregate index and leaf area index. Δ L represents that blade averaging projection area or blade mean breadth are given or use the Wp of Chen&Chilar optimization method by user, Δ A value is the integral multiple of Δ L and is less than image area, and the mean value that aggregate index equals the clearance rate of all statistics community is taken the logarithm divided by the mean value of the logarithm of all communities clearance rate.

${\mathrm{\&Omega;}\left(\mathrm{\&theta;}\right)}_{\mathrm{LX}}=\frac{\ln \left(\mathrm{mean}(P_{\mathrm{cell}}\left(\mathrm{\&theta;}\right)\right)}{\mathrm{mean}\left(\ln \left(P_{\mathrm{cell}}\left(\mathrm{\&theta;}\right)\right)\right)}$

P (θ) and Ω (θ) now_LXAll from two-value classified image, add up and obtained that (these two values P (θ) of first adding up each cell from two-value classified image is P_cell(θ)), had after this parameter and just can utilize above-mentioned formula to calculate Ω (θ)_LX. Suppose that blade tilt obeys in the situation of spherical distribution, G (θ)=0.5, cos (θ) determines according to image angle.

$\mathrm{LAI}=\frac{-\ln \left(P\left(\mathrm{\&theta;}\right)\right)*\cos \left(\mathrm{\&theta;}\right)}{G\left(\mathrm{\&theta;}\right)*\mathrm{\&Omega;}\left(\mathrm{\&theta;}\right)}$

Wherein, θ is zenith angle, and P (θ) is clearance rate, represents that Ray Of Light passes the probability of vegetation canopy along zenith angle θ; G (θ) represents that unit volume leaf area is at the averaging projection's area perpendicular on θ direction plane, relevant with Leaf angle inclination distribution, in the situation that blade tilt is obeyed spherical distribution, and G (θ)=0.5; Ω (θ) is aggregate index, it depends on the spatial distribution of blade, blade regular distribution Ω (θ) > 1, blade random distribution Ω (θ)=1, Leave gathering Ω (θ) < 1. In the situation that Ω (θ) is unknown, the product that can only obtain Ω (θ) and LAI is called again effective leaf area index.

After this step, leaf area index, aggregate index and error under each Δ A are exported.

The second embodiment: based on improved Chen&Chilar method.

Mostly the vegetation canopy of occurring in nature is gathering, and under identical leaf area index condition, the canopy of gathering more may occur large scale gap than the canopy of random distribution. These large-sized gaps have increased canopy clearance rate, and have affected the indirect measurement of leaf area index. Present problem is converted into the maximum gap size of determining that random canopy occurs. The comprehensive canopy clearance rate of this method and gap size information, need to build gap size cumulative distribution in order to answer this problem, the gap size of actual measurement is sorted from small to large, the frequency of each size is cumulative from large scale to small size, measure the clearance rate size integral distribution curve of canopy, the gap size that utilizes gap size removal method to remove nonrandom part from total clearance rate obtains the gap size distribution curve of redistribution, can obtain by this process the key parameter that aggregate index is calculated.

After having told about our ratio juris, then state that in detail routine processes flow process is as follows:

It is the prerequisite of calculating aggregate index that gap size distributes, and this step is most important, is also the innovation place of this method. The process of calculating aggregate index in former method be an actual measurement gap size distribute and stochastic regime under the gap size process of mutually approaching that distributes, but the condition that in former method, iteration stops is very unstable, and parameter Wp cannot automatic acquisition, this has increased the difficulty of application. This method determines by the method for error analysis the condition that Wp and iteration stop. The principle of this process is:

Gap size cumulative distribution under stochastic regime can be utilized the achievement in research of Miller and Norman, they have provided the theoretical formula under horizontal blade sun zenith illuminate condition, and their theory is carried out improved theoretical formula in any leaf inclination angle and solar zenith angle situation by Chen Jingming etc.:

$F\left(\mathrm{\&lambda;}\right)=(1+\frac{L_p*\mathrm{\&lambda;}}{W_p}).\exp (-L_p*(1+\frac{\mathrm{\&lambda;}}{W_p}))$

F (λ) represents that gap size distributes, L_pRepresent the projected area of leaf area on level ground, λ represents gap size, W_pRepresent that former definition blade is in the projection width perpendicular in sunshine direction.

LAICC method is the process of an iteration, and the problem in application is to determine stopping criterion for iteration. W_pGiven according to measuring, L_pA given initial value L_p＝-log(F_m(0)), afterwards by approaching F_m(λ) and two gap sizes of F (λ) each iteration that distributes make L_p＝-log(F_m(0)-Dg)，Dg＝F_m(0)-F_mr(0). Stopping criterion for iteration is L_pTwice growth is less than threshold value, conventionally gets 0.01. Passing threshold is difficult to realize the removal of large gap, the F in order to reach sometimes sometimes in actual applications_m(λ) and F (λ) approach, cause the gap existing under random distribution condition to be also removed (being called removal problem), this does not meet actual conditions. For this problem, propose herein to utilize error analysis to determine the method for stopping criterion for iteration.

The error that the error that leaf area index is calculated equals clearance rate is multiplied by L_pDifferential. Adopt and remove contrary thinking with gap, distribute and build the gap size distribution F (λ) nature from the gap size of random distribution_m. Clearance rate error is cost1, L_pDifferential be cost2, leaf area index calculate error be cost.

$F{{\left(\mathrm{\&lambda;}\right)}^{\&prime;}}_m=(1+\frac{L_p*\mathrm{\&lambda;}}{W_p}).\exp (-L_p*(1+\frac{\mathrm{\&lambda;}}{W_p}))+D_g$

$\cos t1=\sqrt{\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{(F\left(\mathrm{\&lambda;}\right)-\mathrm{Fm}\left(\mathrm{\&lambda;}\right))}^2}{n}}$

F_m(λ) be the clearance rate size integral distribution curve measuring

$\cos t2=\frac{1}{F_m\left(0\right)-\mathrm{Dg}}$

cost＝cost1*cost2

Specific implementation process be exhaustion process for make gap rate variance Dg value (0, (F_m(0) * 0.9)), step-length 0.01; Make W_pValue (1,100), step-length 0.1. The condition that iteration stops is cost while getting minimum of a value, now obtains optimum W_pAnd Dg, utilize this two calculation of parameter aggregate index.

Fig. 9 has shown gap size distribution. In figure, Fm (lambda) represents F_m(λ); F (lambda) represents F (λ); Fmr (lambda) F_mr(λ); When finalF (lambda) final represents iteration termination, Wp and Lp are the gap size under the stochastic regime of inputting.

${\mathrm{\&Omega;}}_E\left(\mathrm{\&theta;}\right)=\frac{\ln \left[F_m\left(0\right)\right]}{\ln \left[F_{\mathrm{mr}}\left(0\right)\right]}\&CenterDot;[1+\frac{F_m\left(0\right)-F_{\mathrm{mr}}\left(0\right)}{1-F_m\left(0\right)}]$

Ω_E(θ) be aggregate index, F_m(λ) be the clearance rate size integral distribution curve measuring, in the time of λ=0, F_m(0) be exactly the canopy clearance rate P (θ) measuring. F (λ) is the gap size integral distribution curve of imaginary canopy leaves spatial stochastically distribution state, F_mr(λ) be F_m(λ) in, remove the gap size distribution curve redistributing behind large scale gap, remove F behind large gap_mr(λ) very approaching with two curves of F (λ), F_mr(0) be the canopy clearance rate of removing behind large scale gap.

3, calculate leaf area index

Calculate leaf area index by leaf area index computing module according to the information parameter extracting.

The first embodiment: based on improved Lang&Xiang method

Determine optimum statistics community area Δ A according to error analysis. Its principle is:

If the average area of blade is Δ L, the area of community is Δ A, when the leaf area index of i community is L_i, and canopy gap distribution is while obeying positive bi-distribution, and the clearance rate of community is:

$P_0={(1-\frac{G}{\mathrm{\&mu;}}\frac{\mathrm{\&Delta;L}}{\mathrm{\&Delta;A}})}^{L_i\mathrm{\&Delta;A}/\mathrm{\&Delta;L}}$

Accordingly, known gap rate P₀Time calculate LAI formula be:

$L_i^{\&prime;}=\frac{\mathrm{\&Delta;L}}{\mathrm{\&Delta;A}}\frac{\ln \left(P_0\right)}{\ln (1-\frac{G}{\mathrm{\&mu;}}\frac{\mathrm{\&Delta;}}{\mathrm{\&Delta;}}\frac{L}{A})}$

If community intra vane is obeyed Poisson distribution completely, LAI formula is:

$L_i^{\&prime;\&prime;}=-\frac{\ln \left(P_0\right)}{\frac{G}{\mathrm{\&mu;}}}$

Visible L_i" with respect to L_i' there is certain over-evaluating, in fact cause that the enchancement factor of LAI evaluated error is a lot, temporarily do not analyze one by one herein, therefore the present invention is simply with L_i″-L_i' as the probabilistic estimation of LAI inverting in community. The error ε of community leaf area index_LiDerive from leaf in community and depart from the error that Poisson distribution causes:

${\mathrm{\&epsiv;}}_{\mathrm{Li}}=L_i^{\&prime;\&prime;}-L_i^{\&prime;}$

$=-\frac{\ln \left(P_0\right)}{\frac{G}{\mathrm{\&mu;}}}-\frac{\mathrm{\&Delta;L}}{\mathrm{\&Delta;A}}\frac{\ln \left(P_0\right)}{\ln (1-\frac{G}{\mathrm{\&mu;}}\frac{\mathrm{\&Delta;L}}{\mathrm{\&Delta;A}})}$

The LAI inverting uncertainty of whole scene is the mean value of each minor family plot error.

${\mathrm{\&epsiv;}}_L=\frac{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{\mathrm{Li}}}{m}$

Δ A during with the reducing leaf area index sigma-delta value and be less than error of cell size is as final community area.

The second embodiment: based on improved Chen&Chilar method.

P (θ) and Ω now_E(θ) all from two-value classified image, add up and obtain, suppose that blade tilt obeys in the situation of spherical distribution, G (θ)=0.5, cos (θ) determines according to image angle.

$\mathrm{LAI}=\frac{-\ln \left(P\left(\mathrm{\&theta;}\right)\right)*\cos \left(\mathrm{\&theta;}\right)}{G\left(\mathrm{\&theta;}\right)*\mathrm{\&Omega;}\left(\mathrm{\&theta;}\right)}$

The maize canopy digital image of the Time Continuous of passing back for data acquisition unit, utilize method of the present invention to extract result LAILX and LAICC estimated value (because the impossible image-position sensor data acquisition of field trial is the same frequent, therefore providing the result contrast of twice field trial herein).

The contrast of table 1LAI hand dipping result

2010-7-8	LAIdest	LAILX	LAICC	LAI-2000
					Liu Zhai	0.51	0.52	0.76	----
Wang Zhuan	1.42	1.28	1.42	----
					Pei Ying	0.54	----	----	----
2010-7-28
					Liu Zhai	3.90	4.31	3.34	3.08
Wang Zhuan	5.53	5.69	5.83	3.11
					Pei Ying	3.85	----	----	3.28

Note: LAIdest: hand dipping, LAILX and LAICC are result LAI2000:LAI-2000 instrument result of the present invention

From table 1 result, LAILX and LAICC result are better than LAI-2000, this be because LAI-2000 measure be effective leaf area index, and other two kinds of methods have been considered aggregate index, what calculate is actual leaf area index, and strict difinition the leaf area index is here full vegetation leaf area index (comprising vegetation cane part). Figure 10 has shown that Hebi, Henan Estimating Leaf Area In Maize index extracts result, think that from Figure 10 the result of two kinds of methods is more approaching, hour resultant error is larger at later stage canopy clearance rate for LAICC method, therefore in the time that canopy clearance rate is too small, does not advise using the method.

Claims (14)

1. an automatic leaf area index observation system, comprising:

Leaf area index automatic Observation server, comprising: bianry image generation module, parameter are carriedDelivery piece, leaf area index computing module, wherein bianry image generation module is used for data acquisitionThe crop canopies digital picture that acquisition means obtains is categorized into bianry image, described parameter extraction moduleFor extracting clearance rate, gap size, aggregate index, described leaf area index from bianry imageComputing module is for calculating leaf area index according to the information parameter extracting;

Data acquisition unit, comprising: for the solar energy intelligent electricity generation system of powering, gather and doThe camera sensing device of thing digital picture, for transmitting the digital transmission module of data and for connecingReceive the data that gather and be transferred to described leaf area index automatic Observation server by digital transmission moduleData acquisition board.

2. according to the automatic leaf area index observation system described in claim l, wherein saidBianry image generation module comprises automatic threshold sort module, described automatic threshold sort module rootEstablish according to the colourity in three wave bands of the red, green, blue to RGB color space and HSL color spaceDetermine threshold value, be green vegetation and Soil Background two classes by Images Classification, thereby transfer binary map toPicture.

3. automatic leaf area index observation system according to claim 1, wherein saidBianry image generation module comprises a point block threshold value sort module, and described point of block threshold value sort module willOriginal image becomes the classification of uniform figure spot, then utilizes wave band threshold value that Images Classification is planted for greenThereby quilt and Soil Background two classes transfer bianry image to.

4. automatic leaf area index observation system according to claim 1, wherein, baseIn improved Lang&Xiang method, described parameter extraction module is determined community area △ A,Under each statistics community area, add up the clearance rate P of each community by cell_cell(θ) itsMiddle △ A value is the integral multiple of △ L and is less than image area, and △ L represents blade averaging projection faceAmass or blade mean breadth.

5. automatic leaf area index observation system according to claim 4, wherein based onImproved Lang&Xiang method, described parameter extraction module is by the gap of all statistics communityThe mean value of rate is taken the logarithm divided by the mean value of the logarithm of all communities clearance rate, calculatesDescribed aggregate index:

$\mathrm{\&Omega;}{\left(\mathrm{\&theta;}\right)}_{LX}=\frac{\ln \left(mean\right(Pcell\left(\mathrm{\&theta;}\right))}{mean\left(ln\right(Pcell\left(\mathrm{\&theta;}\right))}.$

6. automatic leaf area index observation system according to claim 1, wherein based onImproved Chen&Chilar method, in described parameter extraction module statistics digital image in a lineThe pixel count of black pixel, adds up all image lines, then the space to same size continuouslyThe number of times occurring sums up, and obtains the frequency that each size space occurs in entire imageRate, adds and design of graphics picture by descending all gap sizes sequence and by its frequency separatelyGap size distribution curve F_m(λ), λ represents gap size;

Described parameter extraction module according to the gap size obtaining at image statistics and build clearance rulerAfter very little distribution curve, on curve, gap size is that the value of 0 o'clock is clearance rate F_m(0), its etc.Valency is in P (θ);

Described parameter extraction module utilizes error analysis method to determine the Wp of leaf characteristic projection widthCondition with iteration stops, builds the gap size distribution curve of removing large scale gap, F_mr(λ), λ represents gap size, and on curve, gap size is that the value of 0 o'clock is that simulation is randomThe clearance rate F of canopy_mr(0), in conjunction with the described clearance rate F having calculated_m(0) described in calculatingAggregate index:

$\mathrm{\&Omega;}\left(\mathrm{\&theta;}\right)=\frac{\ln \&lsqb;Fm\left(0\right)\&rsqb;\left(mean\right(Pcell\left(\mathrm{\&theta;}\right))}{mean\left(ln\right(Pcell\left(\mathrm{\&theta;}\right))}.$

7. a leaf area index automatic observation process, comprising:

The crop canopies digital picture that data acquisition unit is obtained is categorized into bianry image;

Extract clearance rate, gap size, aggregate index from bianry image;

Calculate leaf area index according to the information parameter extracting.

8. leaf area index automatic observation process according to claim 7, wherein classificationMethod comprises K-average sorting technique, and described K-average sorting technique is according to selected classificationNumber is divided into multiclass automatically, then carries out threshold process according to all kinds of wave band averages, merges into greenVegetation and soil background two classes, thus bianry image transferred to.

9. leaf area index automatic observation process according to claim 7, wherein classificationMethod comprises automatic threshold sorting technique, and described automatic threshold sorting technique basis is to RGB colourColourity setting threshold in three wave bands of red, green, blue and the HSL color space in space, willThereby Images Classification is green vegetation and Soil Background two classes transfers bianry image to.

10. leaf area index automatic observation process according to claim 7, wherein classificationMethod comprises a point block threshold value sorting technique, and described point of block threshold value sorting technique becomes original image allEven figure spot classification, then utilizing wave band threshold value is green vegetation and Soil Background by Images ClassificationThereby two classes transfer bianry image to.

11. leaf area index automatic observation process according to claim 7, based on improvementLang&Xiang method determine community area A, under each statistics community area, press unitLattice are added up the clearance rate P of each community, the integral multiple that wherein A value is L and be less than image surfaceLong-pending, L represents blade averaging projection area or blade mean breadth.

12. leaf area index automatic observation process according to claim 11, based on changingThe Lang&Xiang method of entering, takes the logarithm the mean value of the clearance rate of all statistics community to removeWith the mean value of the logarithm of all communities clearance rate, calculate described aggregate index:

$\mathrm{\&Omega;}{\left(\mathrm{\&theta;}\right)}_{LX}=\frac{\ln \left(mean\right(Pcell\left(\mathrm{\&theta;}\right))}{mean\left(ln\right(Pcell\left(\mathrm{\&theta;}\right))}$

Based on improved Lang&Xiang method, calculate described leaf area according to following formula and refer toNumber:

$LAI=\frac{-\ln \left(P\right(\mathrm{\&theta;}))*Cos\left(\mathrm{\&theta;}\right)}{G\left(\mathrm{\&theta;}\right)*\mathrm{\&Omega;}\left(\mathrm{\&theta;}\right)}$

Wherein G (θ), cos (θ) determine according to image angle.

13. leaf area index automatic observation process according to claim 7, wherein, baseIn improved Chen&Chilar method, black pixel continuously in a line in statistics digital imagePixel count, adds up all image lines, and the number of times then space of same size being occurred carries outAdd and, and obtain the frequency that each size space occurs in entire image, between allThe descending sequence of gap size also adds its frequency separately with the gap size of design of graphics picture and distributesCurve F_m(λ), λ represents gap size.

14. leaf area index automatic observation process according to claim 13, Qi ZhongjiIn improved Chen&Chilar method, utilize error analysis to determine the Wp of leaf characteristic projection widthCondition with iteration stops, builds the gap size distribution curve F that removes large scale gap_mr(λ), λ represents gap size, and on curve, gap size is that the value of 0 o'clock is that simulation is randomThe clearance rate F of canopy_mr(0), in conjunction with the described clearance rate F having calculated_m(0) according to following publicFormula is calculated described aggregate index:

$\mathrm{\&Omega;}\left(\mathrm{\&theta;}\right)=\frac{ln\&lsqb;Fm\left(0\right)\&rsqb;\left(mean\right(Pcell\left(\mathrm{\&theta;}\right))}{mean\left(ln\right(Pcell\left(\mathrm{\&theta;}\right))}.$